CN101576445A - Data reappearing method for structure health monitoring failure sensor simulating memory of people - Google Patents

Abstract

The invention discloses a data reappearance method for anthropomorphic memory structural health monitoring failure sensor: various structural state information collected by the sensor are stored in a long-term memory storage area in time sequence to form a memory image, and the structural state information and memory image are established According to the clue relationship, identify the new structural state information and perform subsequent processing: or, according to the clue relationship, filter the new structural state information to achieve data reduction, or, according to the clue relationship, find the memory corresponding to the new structural state information Mapping, which compensates and reproduces the missing part of the new structural state information. Beneficial technical effects of the present invention: 1) The structural state monitoring information is marked as "old state" or "new state" through the structural state memory identification function, and a large amount of information marked as "old state" can be discarded according to system requirements, thereby realizing structural Data reduction for health monitoring; 2) When a long-term working sensor fails, the monitoring data of the failed sensor can be recovered from the memory bank through the structural state memory recall function, thereby saving the later system maintenance cost.

Description

The structure health monitoring failure sensor data reproduction method of anthropomorphic memory

Technical field

The present invention relates to a kind of microcomputer data processing, relate in particular to a kind of structure health monitoring failure sensor data reproduction method of anthropomorphic memory.

Background technology

Monitoring structural health conditions (Structural Health Monitoring, be called for short SHM) be widely used in great civil engineering works such as complicated mobile device such as aerospace flight vehicle, naval vessel, extra large land vehicle and large bridge, grand building, its task is to rely on sensors such as a large amount of amounts of deflection, strain, displacement, temperature, in real time sensing and evaluation structure health status parameter.The SHM system entails of long-term work can cause operative sensor to lose efficacy because of sensor aging, to the timely discovery of failure sensor and replace that to repair be one of groundwork of SHM system later maintenance.In the actual engineering, hardware is changed and is related to complicated procedures of forming such as installation, debugging.In addition, the sensor that some SHM uses in for example Fibre Optical Sensor has been nuzzled structure in use in advance, is difficult to replace reparation behind the sensor failure.Present a kind of thinking is to utilize software scenario to do the hardware reparation, and the spatial correlation of commonly utilizing data is extrapolated the measurement data of failure sensor by the measuring point of adjacent position, and concrete grammar can adopt data interpolating method or data fusion technology.But said method need be understood CONSTRUCTED SPECIFICATION and set up mathematical model, has certain difficulty on engineering construction.

Memory is the reflection of human brain to past experience, such as the things of past perception, the problem of thinking deeply, the mood of experiencing and emotion, action of doing etc., all may be stored in the brains.It comprises memorize, keeps, re-recognizes and reappears four processes.The time angle that keeps from memory can be divided into: immediate memory, short-term memory, long-term memory.In the step that memory forms, can be divided into following three kinds of information processing manners (the human brain working model is referring to Fig. 1):

1) coding: acquired information is also handled and is made up;

2) store: will make up the information of putting in order and do permanent record;

3) retrieval: the information that is stored is taken out, respond some hints and incident.

The memory retrieval can be divided into memory identification again and memory is recalled.Memory is recalled model and is proposed when studying associative memory model (SAM) by Raaijmakers and Shiffrin, proposes the memory model of cognition by Gillund and Shiffrin on this basis.

In the memory retrieval model, the map information that brain memory is lived is stored in the long-term memory district, and the memory clue is synthetic in the short-term memory district.Every memory clue and map information establish annexation, and each annexation is marked with strength of joint.Memory identification is that all map informations in the information of memory clue guiding and the long-term memory district are comprehensively compared, and differentiates it and is fresh information or old information.It then is a search customizing messages (the memory retrieval model structure of human brain is seen Fig. 2) in all map informations of long-term memory district record that memory is recalled.

Summary of the invention

The present invention proposes a kind of structure health monitoring failure sensor data reproduction method of anthropomorphic memory, this method comprises: the multiple configuration state information that sensor acquisition is arrived, be saved in the long-term memory storage area chronologically and form memory mapping, set up the clue relation of configuration state information and memory mapping; According to the clue relation subsequent treatment is discerned and done to the new construction status information: perhaps, according to the clue relation new construction status information is screened to realize the data yojan, perhaps, concern the memory mapping that finds new construction status information correspondence according to clue, the disappearance in the new construction status information is partly compensated, reappears.

Set up the clue relation of configuration state information and memory mapping according to following formula,

$F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$

In the formula, M is the quantity of memory mapping; N is the quantity of memory clue;

P _j, P _NBeing respectively j kind and the pairing memory clue of N kind configuration state information, is the current data value of j kind and N kind configuration state information also, and j is a numerical sequence, and j=1,2 ... N;

S _kBe k memory mapping of configuration state information; S _MBeing M memory mapping of configuration state information, also is the memory mapping of last position; S _kAnd S _MIn all comprised all original memory clue corresponding historical data values, i.e. the P in the identical moment ₁, P ₂P _NCorresponding configuration state information is kept in the same memory mapping, and k is a numerical sequence, and k=1, and 2 ... M; { S ₁, S ₂, S ₃S _MThe data acquisition formed is the memory mapping storehouse, initiate memory mapping is designated as S _M+1

W _jIt is the weight index;

R (P _j, S _k) be P _jAnd S _kStrength of joint;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value;

S (P _j, S _k[P _j]) be memory clue P _jWith S _k[P _j] the middle P that preserves _jThe relationship degree formula of historical data value.

S (the P that the present invention adopts _j, S _k[P _j]) expression formula is as follows:

$S(P_j,S_k[P_j\left]\right)=\left|\frac{P_j-S_k\left[P_j\right]}{S_k\left[P_j\right]}\right|$

In the formula, P _jBeing the pairing memory clue of j kind configuration state information, also is the current data value of j kind configuration state information;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value;

Method to the new construction status information that collects is discerned comprises: definition new construction status information is the current data value (as strain, pressure, amount of deflection, vibration, temperature etc.) of multiple configuration state information; Current data value P with various configuration state information _j, the clue relational expression F (P of substitution configuration state information and memory mapping ₁, P ₂P _N):

If F (P ₁, P ₂P _N) 〉=θ then is identified as " oldState " with this new construction status information, and promptly this new state information is Already in the memory mapping storehouse;

If F (P ₁, P ₂P _N)＜θ then is identified as " new state " with this new construction status information, this new construction status information need be added in the memory mapping storehouse, forms S _M+1

Wherein, F (P ₁, P ₂P _N) be the clue relational expression of configuration state information and memory mapping information;

θ is set thresholding.

The method that the new construction status information is screened to realize the data yojan is: if a certain new construction status information is identified as " oldState ", then with this new construction status information deletion.

The method that disappearance in the new construction status information is partly compensated, reappears is:

If the current data value of the part-structure status information in a certain new construction status information can't obtain because of sensor failure, and the current data value of another part configuration state information is identified as " oldState ", then according to the pairing memory clue of discernible configuration state information, from the memory mapping storehouse, seek the memory mapping of maximal phase like coupling, according to the memory mapping of the maximal phase that finds the data value that can't obtain because of sensor failure is compensated, reappears like coupling, simultaneously, the pairing P of memory mapping that the maximal phase that finds is seemingly mated _jAnd S _kThe strength of joint fractional increments that adds up makes the time of the memory mapping that search repeatedly visited shorter and shorter.

Compensation, the method for reappearing can be summarized as following steps: be P if the current data of disappearance is worth pairing memory clue 1) _T, then ignore formula $F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$ In S (P _T, S _k[P _T]) its result of back calculating; T is a numerical sequence, and 1≤T≤N;

If F (P ₁, P ₂P _N) satisfy F (P ₁, P ₂P _N) 〉=θ promptly, removes P in the new construction status information _TThe outer pairing current data value of all the other memory clues has been kept in the data base, judges that then the missing data in this new construction status information has reproducibility;

2) have under the prerequisite of reproducibility at missing data, utilize memory to recall the similarity formula and search maximal phase in the data base like the memory mapping of coupling, and with in the memory mapping that finds with memory clue P _TCorresponding S _k[P _T] give memory clue P to current data value disappearance _TThereby, realize compensation, reproduction to missing data;

It is as follows that the similarity formula is recalled in memory:

$J({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$

Wherein, S _kBe k memory mapping of configuration state information;

P _jIt is the pairing memory clue of j kind configuration state information; J is a numerical sequence, and j=1,2 ... N, N is the quantity of memory clue;

W _jIt is the weight index;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value; K is a numerical sequence, and k=1,2 ... M, M are the quantity of memory mapping;

Searching the method that maximal phase seemingly mates memory mapping is:

If it is P that the current data of disappearance is worth pairing memory clue _T, T is a numerical sequence, and 1≤T≤N, then ignores formula $J({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$ In S (P _T, S _k[P _T]), calculate all J (P according to formula ₁, P ₂P _N| S _q), q=1,2 ... M gets minimum J (P ₁, P ₂P _N| S _q) value (minimum J (P ₁, P ₂P _N| S _q) the corresponding memory mapping of value is maximal phase and seemingly mates memory mapping); The corresponding ordinal number of the memory mapping that definition finds is K, and the memory mapping that then finds is S _K[P _T], with S _K[P _T] give memory clue P to current data value disappearance _T

Wherein, S (P _T, S _k[P _T]) be memory clue P _TWith S _k[P _T] the middle P that preserves _TThe relationship degree formula of historical data value;

S _K[P _T] be that the memory clue is P _TConfiguration state information be kept at S _KIn the historical data value; 1≤K≤M

S _qBe q memory mapping of configuration state information, S _qIn comprised memory clue P _TPairing historical data value; Q=1,2 ... M;

To P _jAnd S _kThe add up method of a fractional increments of strength of joint is: by memory clue P ₁, P ₂P _NSearch a certain memory mapping S _kSuccessfully be identified as the memory mapping of maximal phase, then to this memory mapping and corresponding memory clue P like coupling ₁, P ₂P _NThe pairing strength of joint fractional increments v that adds up, that is:

R(P _j，S _k)＝R(P _j，S _k)+v，k＝1，2…M

Wherein, P _jBeing the pairing memory clue of j kind configuration state information, is the current data value of j kind configuration state information also, and j is a numerical sequence, and j=1,2 ... N;

S _kBe k memory mapping of configuration state information; K=1,2 ... M, M are memory mapping quantity.

Useful technique effect of the present invention: 1) the configuration state monitoring information is designated " oldState " or " new state " by configuration state memory recognition function, can abandon in a large number the information that is designated " oldState " according to system's needs, thus the data yojan of implementation structure health monitoring; 2) when the sensor failure of long-term work, recall function by the configuration state memory and from data base, recover the Monitoring Data of failure sensor, thereby saved later stage system maintenance cost.

Description of drawings

Fig. 1, human brain memory working model;

Fig. 2, human brain memory retrieval model;

The memory clue relational model of Fig. 3, the present invention's structure;

Fig. 4, the present invention remember the reproduction flow process;

Fig. 5, strain monitoring experimental platform system are formed photo in kind;

Fig. 6, strain monitoring experiment porch;

Fig. 7, strain bridge circuit theory diagrams

Fig. 8, strain bridge scheme of installation;

Fig. 9, familiarity result of calculation;

Figure 10, new memory mapping data storage figure;

Figure 11, failure sensor data reproduction figure;

Figure 12, the accessed Probability Distribution figure of memory mapping data;

Embodiment

The inventor has proposed a kind of structure health monitoring failure sensor data reproduction method of anthropomorphic memory, the method is the simulating human memory pattern, at the sensor failure problem that occurs in the later maintenance of SHM system, the data rule that model is lived according to memory when certain dimension or several dimension data disappearance is carried out information playback, provide correct assessment result, thereby the monitoring system erroneous judgement of avoiding causing because of the sensor faults itself is broken, and reduces SHM system later maintenance cost simultaneously.This method core is: with the multiple configuration state information that sensor acquisition arrives, be saved in the long-term memory storage area chronologically and form memory mapping, set up the clue relation of configuration state information and memory mapping; According to the clue relation subsequent treatment is discerned and done to the new construction status information: perhaps, according to the clue relation new construction status information is screened to realize the data yojan, perhaps, concern the memory mapping that finds new construction status information correspondence according to clue, the disappearance in the new construction status information is partly compensated, reappears.

Adopt the system of this method can be divided into two working stages: the early stage learning phase, the operation phase in later stage;

Referring to Fig. 3, system is in early stage during learning phase: at a certain works, with sensor acquisition to different types of configuration state information be labeled as P in turn ₁, P ₂P _N, P _NThe memory clue that just is called N kind configuration state information, P _NNumerical value be the current data that sensor acquisition arrives; At T ₁Constantly, with the P that collects ₁, P ₂P _NData are recorded in memory mapping S ₁In, with T ₂The P that constantly collects ₁, P ₂P _NData are recorded in memory mapping S ₂In, mode like this is with the P in the follow-up identical moment ₁, P ₂P _NData are recorded in the memory mapping S of serial number _kIn, after memory mapping quantity forms on a fairly large scale, can think that memory mapping storehouse (data base) basically forms; It is worthy of note, be not that a dirt is constant after data base forms, and also may need to add new memory mapping in the operational process.

Operation phase in later stage: structure is after long-time the use, and the sensor that has is owing to expire fatigue lifetime, or other reason damage, causes the operative sensor can't image data, at this moment, and the P of the sign structural health conditions that collects ₁, P ₂P _NData lack with regard to some, and under most of sensors can also the situation of operate as normal, the data that we can arrive according to the sensor acquisition of operate as normal, search in the data base the similar memory mapping of part normal data therewith, historical data according to corresponding missing data in the memory mapping is reappeared missing data, for the evaluation of structural health conditions provides foundation.Data reproduction is the aspect that the present invention uses, and the another application of the invention aspect is, according to the P that newly collects ₁, P ₂P _NData are searched and whether are preserved the similar memory mapping that similarity meets the demands in the data base, if having then delete the P that newly collects ₁, P ₂P _NData realize the yojan of system data.

During set forth summary of the present invention the front, the Several Key Problems that relates to was: how to set up configuration state information and concern with the clue of memory mapping, how to search memory mapping similar to normal data in (identification) data base, how to judge the similarity of memory mapping and new data; Further describe with regard to above 3 problems below.

(1) memory identification rule:

GillundG and ShiffrinRM have proposed a fairly simple human brain memory familiarity formula:

$F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^{N}R{(P_j,S_k)}^{W_j}$

This formula is by introducing strength of joint R (P _j, S _k) index has been set up between memory clue and memory mapping has the mapping relations that intensity quantizes, for analyst's brain memory function provides the mathematical model basis.Use for reference the characteristics of this technological thought and integrated structure health monitoring, the present invention introduces relationship degree computing function in preceding formula, proposes following memory identification familiarity formula:

$F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$

In the formula, M is the quantity of memory mapping; N is the quantity of memory clue;

P _j, P _NBeing respectively j kind and the pairing memory clue of N kind configuration state information, is the current data value of j kind and N kind configuration state information also, and j is a numerical sequence, and j=1,2 ... N;

S _kBe k memory mapping of configuration state information; S _MBeing M memory mapping of configuration state information, also is the memory mapping of last position; S _kAnd S _MIn all comprised all original memory clue corresponding historical data values, i.e. the P in the identical moment ₁, P ₂P _NCorresponding configuration state information is kept in the same memory mapping, and k is a numerical sequence, and k=1, and 2 ... M; { S ₁, S ₂, S ₃S _MThe data acquisition formed is the memory mapping storehouse, initiate memory mapping is designated as S _M+1

W _jBe reflection memory clue P _jTo memory mapping S _kThe weight index of importance;

R (P _j, S _k) be P _jAnd S _kStrength of joint;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value;

S (P _j, S _k[P _j]) be memory clue P _jWith S _k[P _j] the middle P that preserves _jThe relationship degree formula of historical data value, can be according to the demand structure of SHM application system; S (the P that the present invention adopts _j, S _k[P _j]) expression formula is as follows:

$S(P_j,S_k[P_j\left]\right)=\left|\frac{P_j-S_k\left[P_j\right]}{S_k\left[P_j\right]}\right|$

In the formula, P _jBeing the pairing memory clue of j kind configuration state information, also is the current data value of j kind configuration state information;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value;

To set thresholding θ, memory identification rule is as follows:

1) if F is (P ₁, P ₂P _N) 〉=θ, data processing model is one " oldState " to the configuration state of this memory clue representative of system prompt, Already in the memory mapping storehouse;

2) if F is (P ₁, P ₂P _N)＜θ, data processing model is one " new state " to the configuration state of this memory clue representative of system prompt, needs to add to advance in the memory mapping storehouse, produces memory mapping S _M+1, S _M+1Be at learning phase and finish the new memory mapping (correlating mutually) that adds in the operational process of back with preamble.

Above-mentioned memory identifying is effectively managed the sensor acquisition data, and mass of redundancy data is identified as " oldState " and rejects, and what enter the long-term memory district is the reduced set of memory mapping information.

(2) rule is recalled in memory:

Configuration state is identified as the memory clue P of " oldState " ₁, P ₂P _N, if seek its similar record in the memory mapping storehouse, this process is promptly remembered and is recalled.Consider to remember to recall to come down to a process of seeking maximal phase like occurrence, the present invention proposes following memory and recalls the similarity formula:

$J({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$

Wherein, S _kBe k memory mapping of configuration state information;

P _jIt is the pairing memory clue of j kind configuration state information; J is a numerical sequence, and j=1,2 ... N, N is the quantity of memory clue;

W _jIt is the weight index;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value; K is a numerical sequence, and k=1,2 ... M, M are the quantity of memory mapping;

Searching the method that maximal phase seemingly mates memory mapping is:

If it is P that the current data of disappearance is worth pairing memory clue _T, T is a numerical sequence, and 1≤T≤N, then ignores formula $J({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$ In S (P _T, S _k[P _T]), calculate all J (P according to formula ₁, P ₂P _N| S _q), q=1,2 ... M gets minimum J (P ₁, P ₂P _N| S _q) value; The corresponding ordinal number of the memory mapping that definition finds is K, and the memory mapping that then finds is S _K[P _T], with S _K[P _T] give memory clue P to current data value disappearance _T

Wherein, S (P _T, S _k[P _T]) be memory clue P _TWith S _k[P _T] the middle P that preserves _TThe relationship degree formula of historical data value;

S _K[P _T] be that the memory clue is P _TConfiguration state information be kept at S _KIn the historical data value; 1≤K≤M

S _qBe q memory mapping of configuration state information, S _qIn comprised memory clue P _TPairing historical data value; Q=1,2 ... M;

The human brain memory theory thinks that repetitive learning can enhance memory, sets each other off therewith, and the present invention is by promoting P _jAnd S _kStrength of joint R (P _j, S _k) value simulates this memory characteristic: recall by once successful configuration state memory, find the memory mapping S of maximal phase like coupling _kWith memory clue P ₁, P ₂P _NBe complementary, then be provided with:

R(P _j，S _k)＝R(P _j，S _k)+v，k＝1，2…M

Wherein, P _jBeing the pairing memory clue of j kind configuration state information, is the current data value of j kind configuration state information also, and j is a numerical sequence, and j=1,2 ... N;

S _kBe k memory mapping of configuration state information; K=1,2 ... M, M are memory mapping quantity.

Formula R (P _j, S _k)=R (P _j, S _kThe implication of)+v is: in observation process when long, if the data of certain memory mapping are recalled (recalling) continually, strength of joint between this memory mapping and the memory clue will be enhanced, this means that this memory mapping recalls and have more probability preferentially to be wandered back in the process in memory next time, also promptly reduce search time.According to probability theory, S _kThe chance probability of preferentially being recalled can calculate by following formula:

$P\left(S_k\right|{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\frac{{\mathrm{\&Pi;}}_{j=1}^{N}R{(P_j,S_k)}^{W_j}}{\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^{N}R{(P_j,S_k)}^{W_j}}$

In the formula, formula

Reflection S _kAnd P ₁, P ₂P _NBetween the strength of joint comprehensive effect.

Value big more, memory mapping data S _kPossiblely more wandered back to by priority access.

(3) failure sensor data reproduction:

Memory identification rule provides the criterion of failure sensor data reproducibilitys, possesses the reproducibility of data through the memory identification decision for the configuration state of " oldState " that is:, can further recall rule by memory and realize data reproduction.With above-mentioned two rules as shown in Figure 4 in conjunction with the data reproduction process that is applied to failure sensor.Its method is as follows:

If the current data value of the part-structure status information in a certain new construction status information can't obtain because of sensor failure, and the current data value of another part configuration state information is identified as " oldState ", then according to the pairing memory clue of discernible configuration state information, from the memory mapping storehouse, seek the memory mapping of maximal phase, the data value that can't obtain because of sensor failure is compensated, reappears according to the memory mapping of the maximal phase that finds like coupling like coupling.

Specific practice is:

1) be P if the current data of disappearance is worth pairing memory clue _T, then ignore formula $F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$ In S (P _T, S _k[P _T]) its result of back calculating; T is a numerical sequence, and 1≤T≤N;

If F (P ₁, P ₂P _N) satisfy F (P ₁, P ₂P _N) 〉=θ promptly, removes P in the new construction status information _TThe outer pairing current data value of all the other memory clues has been kept in the data base, judges that then the missing data in this new construction status information has reproducibility;

2) have under the prerequisite of reproducibility at missing data, utilize memory to recall the similarity formula and search maximal phase in the data base like the memory mapping of coupling, and with in the memory mapping that finds with memory clue P _TCorresponding S _k[P _T] give memory clue P to current data value disappearance _TThereby, realize compensation, reproduction to missing data;

Further set forth method of the present invention below in conjunction with embodiment.

Embodiment:

A free beam strain monitoring experiment porch, experimental system is as shown in Figure 5; The experiment porch main body by

1200mm strides the little bent steel plate and the moment of torsion force application apparatus in footpath and forms, equidistant six groups of strain measurement full-bridge circuits that distribute on little bent steel plate based on the metal forming resistance strain gage, and numbered sequence is 6,4,2,1,3,5 from left to right.Turn to pressure and act on little bent steel plate center by the torque axis that will manually apply, cause the steel plate structure strain, see shown in Figure 6.The strain of each monitoring location of foil gauge full-bridge circuit sensing also uploads to host computer through data collector and is for data processing.Data collector adopts NI CRIO of company 9237 and embedded controller CRIO-9014, is responsible for gathering and handles obtaining strain value.Upload to industrial computer through Ethernet interface and carry out the data analysis processing.Each strain bridge is typical Huygens's full-bridge circuit, and circuit theory and installation diagram are seen shown in Fig. 7,8.

Corresponding with memory models, the function class of NI CRIO-9237 is like immediate memory, and the structural strain state is converted to electric signal by the collection of full-bridge strainometer, and CRIO-9237 gathers this signal and passes to controller CRIO-9014 as the immediate memory memory block.After this, controller CRIO-9014 is strain data as the short-term memory district in the memory models with the electric signal conversion processing and the data combination of diverse location is the memory clue.Industrial computer is forever preserved useful information as the long-term memory district in the memory models and is carried out works of treatment such as configuration state data reproduction.

The experiment of configuration state data reproduction is divided into two parts: configuration state memory identification and configuration state memory are recalled.Degenerative process by driving torque handle model configuration.Because free beam is to have flexible steel, so during the revolution of moment of torsion handle, free beam can return to its original state.This specific character makes whole experiment have repeatability.We are at first to structure memory input priori, and then carry out configuration state memory identification and process is recalled in memory.

(1) configuration state memory identification experiment

At first, obtain basic data memory by a learning process.The moment of torsion handle is by Rotate 180 degree successively clockwise, 7 times like this.Six groups of strain datas of a sequence are stored the into long-term memory district of industrial computer, are shown in Table 1.

Table 1

This experiment is with relationship degree formula S (P _j, S _k[P _j]) be defined as follows:

$S(P_j,S_k[P_j\left]\right)=\left|\frac{P_j-S_k\left[P_j\right]}{S_k\left[P_j\right]}\right|$

If P _j=S _k[P _j], S (P then is set _j, S _k[P _j])=10 ^-6In addition, all R (P _j, S _k) initial value all be made as 0.1, k=1,2 ... M; W _j(j=1,2 ... N) all be made as 1.

Simulated data input memory mapping S with the memory clue _T1, S _T2... S _T8And substitution formula $F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$ To determine familiarity thresholding θ, obtain following memory mapping sequence:

S _t1＝[1 50?16 130?50?30]*(-1.0000e-006)

S _t2＝[27?30?147?26 23?30]*(-1.0000e-006)

S _t3＝[27?41?147?30 23?25]*(-1.0000e-006)

S _t4＝[27?41?166?40 25?38]*(-1.0000e-006)

S _t5＝[27?41?166?150?26?40]*(-1.0000e-006)

S _t6＝[27?41?166?150?57?40]*(-1.0000e-006)

S _t7＝[27?41?166?150?57?18]*(-1.0000e-006)

S _t8＝[26?42?164?151?58?20]*(-1.0000e-006)

With the memory mapping S in the table 1 ₄In the memory clue and the memory clue in the above-mentioned memory mapping compare: S _T1And S ₄Different fully, S _T2A clue element and S are arranged ₄The same, S _T3Two clue elements and S are arranged ₄The same, S _T4, S _T5, S _T6, S _T7By that analogy.S _T8Each clue element all and S ₄Approaching.

The result of calculation of familiarity is seen Fig. 9, and as shown in the figure, the memory clue has been divided into different grades with familiarity between memory mapping, and this shows certain particular system, can distinguish the familiarity grade by defining a thresholding θ.

This experiment is provided with thresholding θ=20.After this with the moment of torsion handle from 0 the degree rotate to 180 * n (n=1,2 ...) and the degree.To rotation each time, the data reproduction model can both be that " oldState " or " new state " judge to the memory mapping data according to thresholding indicating structure strain regime.

For example, when the moment of torsion handle is transferred to 180 * 3 when spending, the clue data in the memory mapping are [23 32 148 129 4312] * (1.0000e-006), familiarity result of calculation F (P ₁, P ₂P _N)=3935＞θ, this shows that this strain regime is one " oldState " once took place; When the moment of torsion handle is transferred to 180 * 9 when spending, the clue data in the memory mapping are [47 103 268,251 118 38] * (1.0000e-006), F (P ₁, P ₂P _NIt is one " new state " that)=0.0059＜θ, data reproduction model identify this strain regime thus, should store in the long-term memory district.New memory mapping data storage is seen shown in Figure 10.

(2) experiment is recalled in the configuration state memory

The memory clue that is identified as " oldState " can be recalled its matched data of searching in memory mapping storage data by the configuration state memory.For example, be [23 32 148 129 43 12] * (1.0000e-006), application of formula to memory clue data $J({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2},$ Similarity result of calculation J (P ₁, P ₂P _N| S _k)=[7.9763 1.6262 0.1880 0.8710 1.3751 1.7296 1.9918 2.3833].Because min (J (P ₁, P ₂P _N| S _k))=0.1880, so the memory mapping S that wanders back to _kSequence number be 3.This memory clue data and memory mapping data S accurately ₃=[24 30 147 130 44 13] * is (1.0000e-006) close.This shows that configuration state memory recalls model and can be used as to disturb and suppress to filter, even the Monitoring Data of obtaining has certain noise, recalls process by memory and can eliminate noise effect.

(3) configuration state failure sensor data reproduction experiment

But utilize the memory identification of configuration state and the data recovery that function implementation structure monitors failure sensor is recalled in memory.For example, strainometer 4 and 1 test plug forward the moment of torsion handle to about 180 * 9 positions in pulling out Fig. 6, and * (1.0000e-006) for [46 X, 268 X 119 38] for the memory clue data of obtaining.If P ₂=P ₄=0, by formula $F({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$ Calculate S (P ₂, S _k[P ₂])=1, S (P ₄, S _k[P ₄])=1, F (P ₁, P ₂P _N)=56＞θ, clue is imperfect although this shows memory, and it is one " oldState " for memory mapping, has reproducibility.By formula $J({\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="A20091010401000162.png,A20091010401000192.png"\; state="\{\{state\}\}">P</figure-callout>}}_1,{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="A20091010401000192.png,A20091010401000193.png"\; state="\{\{state\}\}">P</figure-callout>}}_2...P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$ Can calculate J (P ₁, P ₂P _N| S _k)=[14.8458 8.0691 5.3024 3.8335 3.1072 2.6952 2.4910 2.3577].Because min (J (P ₁, P ₂P _N| S _k))=2.3577, therefore remember the memory mapping S that wanders back to _kSequence number be 8.Can from the memory mapping storehouse, find in view of the above and lack big data value P ₂=S ₈[P ₂]=103, P ₄=S ₈[P ₄]=251.The data of acquired original and data recovered are seen shown in Figure 11.

Above-mentioned experiment supposition R (P _j, S _k) (k=1,2 ... value M) is a constant basis.If consider the enhance memory strength of joint, formula R (P can be set _j, S _k)=R (P _j, S _kFractional increments v=0.001 among the)+v is as memory mapping data S ₃Recalled and visited when reaching 10 times memory mapping data S _k(k=1,2 ... 8) when being visited once more by priority access to Probability Distribution see Figure 12.From Figure 12 as seen, by recalled the visit 10 times after, S ₃Have higher probability preferentially to be wandered back to, this phenomenon and human brain memory characteristic are approximate.

Claims (9)

1, a kind of structure health monitoring failure sensor data reproduction method of anthropomorphic memory, it is characterized in that: the multiple configuration state information that sensor acquisition is arrived, be saved in the long-term memory storage area chronologically and form memory mapping, set up the clue relation of configuration state information and memory mapping; According to the clue relation subsequent treatment is discerned and done to the new construction status information: perhaps, according to the clue relation new construction status information is screened to realize the data yojan, perhaps, concern the memory mapping that finds new construction status information correspondence according to clue, the disappearance in the new construction status information is partly compensated, reappears.

2, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 1 is characterized in that: the clue of setting up configuration state information and memory mapping according to following formula concerns,

$F(P_1,P_2\&CenterDot;\&CenterDot;\&CenterDot;P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$

In the formula, M is the quantity of memory mapping; N is the quantity of memory clue;

P _j, P _NBeing respectively j kind and the pairing memory clue of N kind configuration state information, is the current data value of j kind and N kind configuration state information also, and j is a numerical sequence, and j=1,2 ... N;

S _kBe k memory mapping of configuration state information; S _MBeing M memory mapping of configuration state information, also is the memory mapping of last position; S _kAnd S _MIn all comprised all original memory clue corresponding historical data values, i.e. the p in the identical moment ₁, P ₂P _NCorresponding configuration state information is kept in the same memory mapping, and k is a numerical sequence, and k=1, and 2 ... M; { S ₁, S ₂, S ₃S _MThe data acquisition formed is the memory mapping storehouse, initiate memory mapping is designated as S _M+1

W _jIt is the weight index;

R (P _j, S _k) be P _jAnd S _kStrength of joint;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value;

S (P _i, S _k[P _j]) be memory clue P _iWith S _k[P _j] the middle P that preserves _jThe relationship degree formula of historical data value.

3, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 2 is characterized in that: S (P _j, S _k[P _j]) expression formula is as follows:

$S(P_j,S_k[P_j\left]\right)=\left|\frac{P_j-S_k\left[P_j\right]}{S_k\left[P_j\right]}\right|$

In the formula, P _jBeing the pairing memory clue of j kind configuration state information, also is the current data value of j kind configuration state information;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value;

4, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 1, it is characterized in that: the method to the new construction status information that collects is discerned comprises:

Definition new construction status information is the current data value of multiple configuration state information;

Current data value P with various configuration state information _j, the clue relational expression F (P of substitution configuration state information and memory mapping ₁, P ₂P _N):

If F (P ₁, P ₂P _N) 〉=θ then is identified as " oldState " with this new construction status information, and promptly this new state information is Already in the memory mapping storehouse;

If F (P ₁, P ₂P _N)＜θ then is identified as " new state " with this new construction status information, this new construction status information need be added in the memory mapping storehouse, forms S _M+1

Wherein, F (P ₁, P ₂P _N) be the clue relational expression of configuration state information and memory mapping information;

θ is set thresholding.

5, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 1 is characterized in that: the method that the new construction status information is screened to realize the data yojan is:

If a certain new construction status information is identified as " oldState ", then with this new construction status information deletion.

6, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 1, it is characterized in that: the method that the disappearance in the new construction status information is partly compensated, reappears is:

If the current data value of the part-structure status information in a certain new construction status information can't obtain because of sensor failure, and the current data value of another part configuration state information is identified as " oldState ", then according to the pairing memory clue of discernible configuration state information, from the memory mapping storehouse, seek the memory mapping of maximal phase like coupling, according to the memory mapping of the maximal phase that finds the data value that can't obtain because of sensor failure is compensated, reappears like coupling, simultaneously, the pairing P of memory mapping that the maximal phase that finds is seemingly mated _jAnd S _kThe strength of joint fractional increments that adds up makes the time of the memory mapping that search repeatedly visited shorter and shorter.

7, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 6 is characterized in that: be P if the current data of disappearance is worth pairing memory clue 1) _T, then ignore formula $F(P_1,P_2\&CenterDot;\&CenterDot;\&CenterDot;P_N)=\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&Pi;}}_{j=1}^N\frac{R{(P_j,S_k)}^{W_j}}{S(P_j,S_k[P_j\left]\right)}$ In S (P _T, S _k[P _T]) its result of back calculating; T is a numerical sequence, and 1≤T≤N;

If F (P ₁, P ₂P _N) satisfy F (P ₁, P ₂P _N) 〉=θ promptly, removes P in the new construction status information _TThe outer pairing current data value of all the other memory clues has been kept in the data base, judges that then the missing data in this new construction status information has reproducibility;

2) have under the prerequisite of reproducibility at missing data, utilize memory to recall the similarity formula and search maximal phase in the data base like the memory mapping of coupling, and with in the memory mapping that finds with memory clue P _TCorresponding S _k[P _T] give memory clue P to current data value disappearance _TThereby, realize compensation, reproduction to missing data;

8, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 7 is characterized in that: it is as follows that the similarity formula is recalled in memory:

$J(P_1,P_2\&CenterDot;\&CenterDot;\&CenterDot;P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$

Wherein, S _kBe k memory mapping of configuration state information;

P _jIt is the pairing memory clue of j kind configuration state information; J is a numerical sequence, and j=1,2 ... N, N is the quantity of memory clue;

W _jIt is the weight index;

S _k[P _j] be that the memory clue is P _jConfiguration state information be kept at S _kIn the historical data value; K is a numerical sequence, and k=1,2 ... M, M are the quantity of memory mapping;

Searching the method that maximal phase seemingly mates memory mapping is:

If it is P that the current data of disappearance is worth pairing memory clue _T, T is a numerical sequence, and 1≤T≤N, then ignores formula $J(P_1,P_2\&CenterDot;\&CenterDot;\&CenterDot;P_N|S_k)=\frac{1}{N}\sqrt{\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left[\frac{S(P_j,S_k[P_j\left]\right)}{R{(P_j,S_k)}^{W_j}}\right]}^2}$ In S (P _T, S _k[P _T]), calculate all J (P according to formula ₁, P ₂P _N| S _q), q=1,2 ... M gets minimum J (P ₁, P ₂P _N| S _q) value; The corresponding ordinal number of the memory mapping that definition finds is K, and the memory mapping that then finds is S _K[P _T], with S _K[P _T] give memory clue P to current data value disappearance _T

Wherein, S (P _T, S _k[P _T]) be memory clue P _TWith S _k[P _T] the middle P that preserves _TThe relationship degree formula of historical data value;

S _K[P _T] be that the memory clue is P _TConfiguration state information be kept at S _KIn the historical data value; 1≤K≤M

S _qBe q memory mapping of configuration state information, S _qIn comprised memory clue P _TPairing historical data value; Q=1,2 ... M;

9, the structure health monitoring failure sensor data reproduction method of anthropomorphic memory according to claim 6 is characterized in that: to P _jAnd S _kThe strength of joint fractional increments that adds up comprises:

By memory clue P ₁, P ₂P _NSearch a certain memory mapping S _kSuccessfully be identified as the memory mapping of maximal phase, then to this memory mapping and corresponding memory clue P like coupling ₁, P ₂P _NThe pairing strength of joint fractional increments v that adds up, that is:

R(P _j，S _k)＝R(P _j，S _k)+v，k＝1，2…M

Wherein, P _jBeing the pairing memory clue of j kind configuration state information, is the current data value of j kind configuration state information also, and j is a numerical sequence, and j=1,2 ... N;

S _kBe k memory mapping of configuration state information; K=1,2 ... M, M are memory mapping quantity.

Images